Reflex: Fast Robot Policy Serving by Reusing and Correcting Past Actions
Reflex: Fast Robot Policy Serving by Reusing and Correcting Past Actions
This research presents Reflex, a new serving layer for robot policies that addresses latency and throughput bottlenecks by caching complete action chunks rather than intermediate computations. The method employs a two-level correction mechanism to handle observation-action mismatches and accumulated errors without modifying the underlying policy. Evaluated on representative workloads and real robots, Reflex improves inference latency by 1.4x to 11x and throughput by 2.0x to 2.2x while preserving comparable task success rates.
The paper introduces Reflex, a serving layer that caches and corrects complete action chunks to bypass the expensive policy backbone, achieving 1.4x to 11x latency improvements and 2.0x to 2.2x throughput gains on real robots while maintaining task success rates comparable to baseline methods.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
inference latency improvement
1.4
View original evidence
isions, assembles compatible chunks, and applies lightweight residual correction to limit action mismatch. At the execution level, it captures task progress and adaptively refreshes the original policy to limit accumulated errors and balance inference cost. Without modifying the underlying policy, Reflex improves inference latency by 1.4x to 11x and throughput by 2.0x to 2.2x on representative workloads and real robo
Open source S5inference latency improvement
11
View original evidence
isions, assembles compatible chunks, and applies lightweight residual correction to limit action mismatch. At the execution level, it captures task progress and adaptively refreshes the original policy to limit accumulated errors and balance inference cost. Without modifying the underlying policy, Reflex improves inference latency by 1.4x to 11x and throughput by 2.0x to 2.2x on representative workloads and real robo
Open source S5throughput improvement
2
View original evidence
isions, assembles compatible chunks, and applies lightweight residual correction to limit action mismatch. At the execution level, it captures task progress and adaptively refreshes the original policy to limit accumulated errors and balance inference cost. Without modifying the underlying policy, Reflex improves inference latency by 1.4x to 11x and throughput by 2.0x to 2.2x on representative workloads and real robo
Open source S5throughput improvement
2.2
View original evidence
isions, assembles compatible chunks, and applies lightweight residual correction to limit action mismatch. At the execution level, it captures task progress and adaptively refreshes the original policy to limit accumulated errors and balance inference cost. Without modifying the underlying policy, Reflex improves inference latency by 1.4x to 11x and throughput by 2.0x to 2.2x on representative workloads and real robo
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
isions, assembles compatible chunks, and applies lightweight residual correction to limit action mismatch. At the execution level, it captures task progress and adaptively refreshes the original policy to limit accumulated errors and balance inference cost. Without modifying the underlying policy, Reflex improves inference latency by 1.4x to 11x and throughput by 2.0x to 2.2x on representative workloads and real robo
Open source S5
o avoid redundant computation. Surprisingly, we find that such reuse often offers little advantage over simply running fewer denoising steps from fresh noise, without any cache. Moreover, accelerating denoising leaves the expensive policy backbone intact, which can account for 70% of GPU time. We present Reflex, a serving layer that turns past policy executions into a backbone-free path for future decisions. Rather t
Open source S3
ts, while preserving comparable task success.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org